I have a phonelog table that has information about callers' call history. I'd like to find out callers whose first and last call was to the same person on a given day.
Callerid Recipientid DateCalled
1 2 2019-01-01 09:00:00.000
1 3 2019-01-01 17:00:00.000
1 4 2019-01-01 23:00:00.000
2 5 2019-07-05 09:00:00.000
2 5 2019-07-05 17:00:00.000
2 3 2019-07-05 23:00:00.000
2 5 2019-07-06 17:00:00.000
2 3 2019-08-01 09:00:00.000
2 3 2019-08-01 17:00:00.000
2 4 2019-08-02 09:00:00.000
2 5 2019-08-02 10:00:00.000
2 4 2019-08-02 11:00:00.000
Expected Output
Callerid Recipientid Datecalled
2 5 2019-07-05
2 3 2019-08-01
2 4 2019-08-02
I wrote the below query but can't get it to return recipientid. Any help on this will be appreciated!
select pl.callerid,cast(pl.datecalled as date) as datecalled
from phonelog pl inner join (select callerid, cast(datecalled as date) as datecalled,
min(datecalled) as firstcall, max(datecalled) as lastcall
from phonelog
group by callerid, cast(datecalled as date)) as x
on pl.callerid = x.callerid and cast(pl.datecalled as date) = x.datecalled
and (pl.datecalled = x.firstcall or pl.datecalled = x.lastcall)
group by pl.callerid, cast(pl.datecalled as date)
having count(distinct recipientid) = 1
Another dbFiddle option
First, my prequery (PQ alias), I am getting for a given client, per day, the min and max time called but also HAVING to make sure person had at least 2 phone calls in a given day. From that, I re-join to the phone log table on the FIRST (MIN) call for the person for the given day. Then I join one more time for the LAST (MAX) call for the same person for the same day and make sure the recipient of the first is same as last.
I do not have to join on the stripped-down "JustDate" column used for the grouping as the MIN/MAX qualifies the FULL date/time.
select
PQ.JustDate,
PQ.CallerID,
pl1.RecipientID
from
( select
callerID,
convert( date, dateCalled ) JustDate,
min( DateCalled ) minDateCall,
max( DateCalled ) maxDateCall
from
PhoneLog pl
group by
callerID,
convert( date, dateCalled )
having
count(*) > 1) PQ
JOIN PhoneLog pl1
on PQ.CallerID = pl1.CallerID
AND PQ.minDateCall = pl1.dateCalled
JOIN PhoneLog pl2
on PQ.CallerID = pl2.CallerID
AND PQ.maxDateCall = pl2.dateCalled
AND pl1.RecipientID = pl2.RecipientID
Its very easy with window function
WITH cte AS (
SELECT *, CAST(DateCalled as DATE) DateCalled
,FIRST_VALUE(Recipientid) OVER (PARTITION BY Callerid ,CAST(DateCalled as date) ORDER BY CAST(DateCalled AS DATE)) f
,LAST_VALUE(Recipientid) OVER (PARTITION BY Callerid ,CAST(DateCalled as date) ORDER BY CAST(DateCalled AS DATE)) l
FROM phonelog
)
SELECT DISTINCT Callerid,Recipientid, DateCalled FROM cte
WHERE f=l
Since SQL Server 2019 you could use the first_value() and last_value() window functions.
SELECT DISTINCT
x1.callerid,
x1.fri,
x1.datecalled
FROM (SELECT pl1.callerid,
pl1.recipientid,
convert(date, pl1.datecalled) datecalled,
first_value(pl1.recipientid) OVER (PARTITION BY pl1.callerid,
convert(date, pl1.datecalled)
ORDER BY pl1.datecalled
RANGE BETWEEN UNBOUNDED PRECEDING
AND UNBOUNDED FOLLOWING) fri,
last_value(pl1.recipientid) OVER (PARTITION BY pl1.callerid,
convert(date, pl1.datecalled)
ORDER BY pl1.datecalled
RANGE BETWEEN UNBOUNDED PRECEDING
AND UNBOUNDED FOLLOWING) lri
FROM phonelog pl1) x1
WHERE x1.fri = x1.lri;
In older versions you can use correlated subqueries with TOP 1.
SELECT DISTINCT
x1.callerid,
x1.fri,
x1.datecalled
FROM (SELECT pl1.callerid,
pl1.recipientid,
convert(date, pl1.datecalled) datecalled,
(SELECT TOP 1
pl2.recipientid
FROM phonelog pl2
WHERE pl2.callerid = pl1.callerid
AND pl2.datecalled >= convert(date, pl1.datecalled)
AND pl2.datecalled < dateadd(day, 1, convert(date, pl1.datecalled))
ORDER BY pl2.datecalled ASC) fri,
(SELECT TOP 1
pl2.recipientid
FROM phonelog pl2
WHERE pl2.callerid = pl1.callerid
AND pl2.datecalled >= convert(date, pl1.datecalled)
AND pl2.datecalled < dateadd(day, 1, convert(date, pl1.datecalled))
ORDER BY pl2.datecalled DESC) lri
FROM phonelog pl1) x1
WHERE x1.fri = x1.lri;
db<>fiddle
If you don't want to return log rows where somebody just made one call on a day, which of course means the first and the last call of the day were to the same person, you can use GROUP BY and HAVING count(*) > 1 instead of DISTINCT.
SELECT x1.callerid,
x1.fri,
x1.datecalled
FROM (...) x1
WHERE x1.fri = x1.lri
GROUP BY x1.callerid,
x1.fri,
x1.datecalled
HAVING count(*) > 1;
You can use a CTE to compute the first and last call of each day by Callerid, and then self-JOIN that CTE to find callers whose first and last calls were to the same Recipientid:
WITH CTE AS (
SELECT Callerid, RecipientId, CONVERT(DATE, Datecalled) AS Datecalled,
ROW_NUMBER() OVER (PARTITION BY Callerid, CONVERT(DATE, Datecalled) ORDER BY Datecalled) AS rna,
ROW_NUMBER() OVER (PARTITION BY Callerid, CONVERT(DATE, Datecalled) ORDER BY Datecalled DESC) AS rnb
FROM phonelog
)
SELECT c1.Callerid, c1.RecipientId, c1.Datecalled
FROM CTE c1
JOIN CTE c2 ON c1.Callerid = c2.Callerid AND c1.Recipientid = c2.Recipientid
WHERE c1.rna = 1 AND c2.rnb = 1
Output:
Callerid RecipientId Datecalled
2 5 2019-07-05
2 3 2019-08-01
2 4 2019-08-02
Demo on SQLFiddle
As my understanding, you want to select callerid with each Recipientid with the times greater than 1 to make sure that we have First call and Last call. So you just need to group by 3 columns combine with having count(Recipientid) > 1 Like this
SELECT Callerid, Recipientid, CAST(Datecalled AS DATE) AS Datecalled
FROM phonelog
GROUP BY Callerid, Recipientid, CAST(Datecalled AS DATE)
HAVING COUNT(Recipientid) > 1
Demo on db<>fiddle
As per my understanding we have to rank Caller_id as well as Recipient_id along with the Date.
Below is my solution which is working well for this case.
with CTE as
(select *,
row_number() over (partition by callerid, convert(VARCHAR,datecalled,23) order by convert(VARCHAR,datecalled,23)) as first_recipient_id,
row_number() over (partition by receipientid, convert(VARCHAR,datecalled,23) order by convert(VARCHAR,datecalled,23) desc) as last_recipient_id
from activity
)
select t.callerid,t.receipientid,CONVERT(VARCHAR,t.datecalled) as DateCalled from CTE t
where t.first_recipient_id >1 AND t.last_recipient_id>1;
The result that I was able to get:
Result
I think we need to identify first and last call made by caller on a day and then compare it with first and last call by caller to a recipient for that day. Below code has firstcall and lastcall made by caller on a day. Then it finds first and last call by caller to respective recipient and then compare.
SELECT DISTINCT
callerid,
recipientid,
CONVERT(date,firstcall)
FROM
(
Select
callerid,
recipientid,
MIN(dateCalled) OVER(PARTITION BY callerid,CONVERT(date,DateCalled)) as firstcall,
MAX(DateCalled) OVER(PARTITION BY callerid,CONVERT(date,DateCalled)) as lastcall,
MIN(DateCalled) OVER(PARTITION BY callerid,recipientid,convert(date,DateCalled)) as recipfirstcall,
MAX(call_start_time) OVER(PARTITION BY callerid,recipientid,convert(date,DateCalled)) as reciplastcall
from phonelog
) as A
where A.firstcall=A.recipfirstcall and A.lastcall=A.reciplastcall
Related
Here is an example:
Id|price|Date
1|2|2022-05-21
1|3|2022-06-15
1|2.5|2022-06-19
Needs to look like this:
Id|Date|price
1|2022-05-21|2
1|2022-05-22|2
1|2022-05-23|2
...
1|2022-06-15|3
1|2022-06-16|3
1|2022-06-17|3
1|2022-06-18|3
1|2022-06-19|2.5
1|2022-06-20|2.5
...
Until today
1|2022-08-30|2.5
I tried using the lag(price) over (partition by id order by date)
But i can't get it right.
I'm not familiar with Azure, but it looks like you need to use a calendar table, or generate missing dates using a recursive CTE.
To get started with a recursive CTE, you can generate line numbers for each id (assuming multiple id values) in the source data ordered by date. These rows with row number equal to 1 (with the minimum date value for the corresponding id) will be used as the starting point for the recursion. Then you can use the DATEADD function to generate the row for the next day. To use the price values from the original data, you can use a subquery to get the price for this new date, and if there is no such value (no row for this date), use the previous price value from CTE (use the COALESCE function for this).
For SQL Server query can look like this
WITH cte AS (
SELECT
id,
date,
price
FROM (
SELECT
*,
ROW_NUMBER() OVER (PARTITION BY id ORDER BY date) AS rn
FROM tbl
) t
WHERE rn = 1
UNION ALL
SELECT
cte.id,
DATEADD(d, 1, cte.date),
COALESCE(
(SELECT tbl.price
FROM tbl
WHERE tbl.id = cte.id AND tbl.date = DATEADD(d, 1, cte.date)),
cte.price
)
FROM cte
WHERE DATEADD(d, 1, cte.date) <= GETDATE()
)
SELECT * FROM cte
ORDER BY id, date
OPTION (MAXRECURSION 0)
Note that I added OPTION (MAXRECURSION 0) to make the recursion run through all the steps, since the default value is 100, this is not enough to complete the recursion.
db<>fiddle here
The same approach for MySQL (you need MySQL of version 8.0 to use CTE)
WITH RECURSIVE cte AS (
SELECT
id,
date,
price
FROM (
SELECT
*,
ROW_NUMBER() OVER (PARTITION BY id ORDER BY date) AS rn
FROM tbl
) t
WHERE rn = 1
UNION ALL
SELECT
cte.id,
DATE_ADD(cte.date, interval 1 day),
COALESCE(
(SELECT tbl.price
FROM tbl
WHERE tbl.id = cte.id AND tbl.date = DATE_ADD(cte.date, interval 1 day)),
cte.price
)
FROM cte
WHERE DATE_ADD(cte.date, interval 1 day) <= NOW()
)
SELECT * FROM cte
ORDER BY id, date
db<>fiddle here
Both queries produces the same results, the only difference is the use of the engine's specific date functions.
For MySQL versions below 8.0, you can use a calendar table since you don't have CTE support and can't generate the required date range.
Assuming there is a column in the calendar table to store date values (let's call it date for simplicity) you can use the CROSS JOIN operator to generate date ranges for the id values in your table that will match existing dates. Then you can use a subquery to get the latest price value from the table which is stored for the corresponding date or before it.
So the query would be like this
SELECT
d.id,
d.date,
(SELECT
price
FROM tbl
WHERE tbl.id = d.id AND tbl.date <= d.date
ORDER BY tbl.date DESC
LIMIT 1
) price
FROM (
SELECT
t.id,
c.date
FROM calendar c
CROSS JOIN (SELECT DISTINCT id FROM tbl) t
WHERE c.date BETWEEN (
SELECT
MIN(date) min_date
FROM tbl
WHERE tbl.id = t.id
)
AND NOW()
) d
ORDER BY id, date
Using my pseudo-calendar table with date values ranging from 2022-05-20 to 2022-05-30 and source data in that range, like so
id
price
date
1
2
2022-05-21
1
3
2022-05-25
1
2.5
2022-05-28
2
10
2022-05-25
2
100
2022-05-30
the query produces following results
id
date
price
1
2022-05-21
2
1
2022-05-22
2
1
2022-05-23
2
1
2022-05-24
2
1
2022-05-25
3
1
2022-05-26
3
1
2022-05-27
3
1
2022-05-28
2.5
1
2022-05-29
2.5
1
2022-05-30
2.5
2
2022-05-25
10
2
2022-05-26
10
2
2022-05-27
10
2
2022-05-28
10
2
2022-05-29
10
2
2022-05-30
100
db<>fiddle here
I have this table below
date
id
value
2021-01-01
1
3
2021-01-04
1
5
2021-01-05
1
10
And I expect output like this, where the date column is always increase daily and value column will generate the last value on an id
date
id
value
2021-01-01
1
3
2021-01-02
1
3
2021-01-03
1
3
2021-01-04
1
5
2021-01-05
1
10
2021-01-06
1
10
I think I can use cross join but I can't get my expected output and think that there are a special syntax/logic to solve this
Consider below approach
select * from `project.dataset.table`
union all
select missing_date, prev_row.id, prev_row.value
from (
select *, lag(t) over(partition by id order by date) prev_row
from `project.dataset.table` t
), unnest(generate_date_array(prev_row.date + 1, date - 1)) missing_date
I would write this using:
select dte, t.id, t.value
from (select t.*,
lead(date, 1, date '2021-01-06') over (partition by id order by date) as next_day
from `table` t
) t cross join
unnest(generate_date_array(
date,
ifnull(
date_add(next_date, interval -1 day), -- generate missing date rows
(select max(date) from `table`) -- add last row
)
)) dte;
Note that this requires neither union all nor window function to fill in the values.
alternative solution using last_value. You may explore the following query and customize your logic to generate days (if needed)
WITH
query AS (
SELECT
date,
id,
value
FROM
`mydataset.newtable`
ORDER BY
date ),
generated_days AS (
SELECT
day
FROM (
SELECT
MIN(date) min_dt,
MAX(date) max_dt
FROM
query),
UNNEST(GENERATE_DATE_ARRAY(min_dt, max_dt)) day )
SELECT
g.day,
LAST_VALUE(q.id IGNORE NULLS) OVER(ORDER BY g.day) id,
LAST_VALUE(q.value IGNORE NULLS) OVER(ORDER BY g.day) value,
FROM
generated_days g
LEFT OUTER JOIN
query q
ON
g.day = q.date
ORDER BY
g.day
I have a data ranges with start and end date for a persons, I want to get the continuous date ranges only per persons:
Input:
NAME | STARTDATE | END DATE
--------------------------------------
MIKE | **2019-05-15** | 2019-05-16
MIKE | 2019-05-17 | **2019-05-18**
MIKE | 2020-05-18 | 2020-05-19
Expected output like:
MIKE | **2019-05-15** | **2019-05-18**
MIKE | 2020-05-18 | 2020-05-19
So basically output is MIN and MAX for each continuous period for the person.
Appreciate any help.
I have tried the below query:
With N AS ( SELECT Name, StartDate, EndDate
, LastStop = MAX(EndDate)
OVER (PARTITION BY Name ORDER BY StartDate, EndDate
ROWS BETWEEN UNBOUNDED PRECEDING AND 1 PRECEDING) FROM Table ), B AS ( SELECT Name, StartDate, EndDate
, Block = SUM(CASE WHEN LastStop Is Null Then 1
WHEN LastStop < StartDate Then 1
ELSE 0
END)
OVER (PARTITION BY Name ORDER BY StartDate, LastStop) FROM N ) SELECT Name
, MIN(StartDate) DateFrom
, MAX(EndDate) DateTo FROM B GROUP BY Name, Block ORDER BY Name, Block
But its not considering the continuous period. It's showing the same input.
This is a type of gap-and-islands problem. There is no need to expand the data out by day! That seems very inefficient.
Instead, determine the "islands". This is where there is no overlap -- in your case lag() is sufficient. Then a cumulative sum and aggregation:
select name, min(startdate), max(enddate)
from (select t.*,
sum(case when prev_enddate >= dateadd(day, -1, startdate) then 0 else 1 end) over
(partition by name order by startdate) as grp
from (select t.*,
lag(enddate) over (partition by name order by startdate) as prev_enddate
from t
) t
) t
group by name, grp;
Here is a db<>fiddle.
Here is an example using an ad-hoc tally table
Example or dbFiddle
;with cte as (
Select A.[Name]
,B.D
,Grp = datediff(day,'1900-01-01',D) - dense_rank() over (partition by [Name] Order by D)
From YourTable A
Cross Apply (
Select Top (DateDiff(DAY,StartDate,EndDate)+1) D=DateAdd(DAY,-1+Row_Number() Over (Order By (Select Null)),StartDate)
From master..spt_values n1,master..spt_values n2
) B
)
Select [Name]
,StartDate= min(D)
,EndDate = max(D)
From cte
Group By [Name],Grp
Returns
Name StartDate EndDate
MIKE 2019-05-15 2019-05-18
MIKE 2020-05-18 2020-05-19
Just to help with the Visualization, the CTE generates the following
This will give you the same result
SELECT subquery.name,min(subquery.startdate),max(subquery.enddate1)
FROM (SELECT NAME,startdate,
CASE WHEN EXISTS(SELECT yt1.startdate
FROM t yt1
WHERE yt1.startdate = DATEADD(day, 1, yt2.enddate)
) THEN null else yt2.enddate END as enddate1
FROM t yt2) as subquery
GROUP by NAME, CAST(MONTH(subquery.startdate) AS VARCHAR(2)) + '-' + CAST(YEAR(subquery.startdate) AS VARCHAR(4))
For the CASE WHEN EXISTS I refered to SQL CASE
For the group by month and year you can see this GROUP BY MONTH AND YEAR
DB_FIDDLE
I Have the following SQL Tables:
[Calendar]
[CalendarId]
[Name]
SAMPLE DATA:
CalendarId ResourceKey Name
1 1 tk1-Room1
2 2 tk1-Room2
3 3 tk1-noentries
[CalendarEntry]
[CalendarId]
[CalendarEntryId]
[Start]
[End]
SAMPLE DATA:
CalendarId Start End
1 2019-11-18 16:00:00.0000000 2019-11-18 17:00:00.0000000
1 2019-11-19 16:00:00.0000000 2019-11-19 17:00:00.0000000
2 2019-11-25 16:00:00.0000000 2019-11-25 17:00:00.0000000
1 2019-11-25 17:00:00.0000000 2019-11-25 18:00:00.0000000
Expected output:
Name StartDate EndDate ResourceKey
tk1-Room1 2019-11-25 17:00:00 2019-11-25 17:00:00 1
tk1-Room2 2019-11-25 16:00:00 2019-11-25 17:00:00 2
tk1-noentries NULL NULL 3
I am trying to list all Calendar entries, with their corresponding Start(Most Recent) and End Times.
I have the following code which is working partially:
SELECT Name,StartDate,ResourceKey FROM [Calendar].[dbo].[Calendar] CAL
LEFT JOIN(
SELECT
CalendarId,
MAX(ENT.[Start]) as StartDate
FROM [CalendarEntry] ENT
GROUP BY CalendarId
)
AS ST on CAL.CalendarId = ST.CalendarId
However, If i was to include that column, In my sub SELECT, EG:
SELECT
CalendarId,
MAX(ENT.[Start]) as StartDate,
ENT.[End] as endDate
I get the following error:
Column 'CalendarEntry.End' is invalid in the select list because it is not contained in either an aggregate function or the GROUP BY clause.
However, including it in the GROUP BY now causes Multiple CalendarEntry Rows to come back for each Calendar..
What is the best way for me to grab the most recent row out of CalendarEntry which allows me access to all the columns?
Thanks!
This is a typical top 1 per group question.
You can either use row_number():
select *
from (
select
c.*,
e.*,
row_number() over(partition by c.CalendarId order by e.Start desc) rn
from [Calendar].[dbo].[Calendar] c
left join [CalendarEntry] e ON c.CalendarId = e.CalendarId
) t
where rn = 1
Or you can filter with a correlated subquery:
select c.*, e.*
from [Calendar].[dbo].[Calendar] c
left join [CalendarEntry] e
on c.CalendarId = e.CalendarId
and c.Start = (
select max(e1.Start) from [CalendarEntry] e where c.CalendarId = e1.CalendarId
)
I am trying to list all Calendar entries, with their corresponding Start(Most Recent) and End Times.
I interpret this as the most recent record from CalendarEntry for each CalendarId:
select ce.*
from CalendarEntry ce
where ce.StartDate = (select max(ce2.StartDate)
from CalendarEntry ce2
where ce2.CalendarId = ce.CalendarId
);
You can try OUTER APPLY too, however #GMB's answer is a better approach from performance prospective
SELECT Name,
StartDate,
EndDate,
ResourceKey
FROM dbo.Calendar AS C
OUTER APPLY
(
SELECT TOP 1 *
FROM dbo.CalendarEntry
WHERE CalendarId = C.CalendarId
ORDER BY StartDate DESC,
EndDate DESC
) AS K;
You can also try LAST_VALUE/FIRST_VALUE(available in SQL Server 2012 and later) functions too, as below, , however again #GMB's answer is a better approach from performance prospective:
SELECT DISTINCT
Name,
LAST_VALUE(StartDate) OVER (PARTITION BY C.CalendarId
ORDER BY StartDate,EndDate
ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING),
LAST_VALUE(EndDate) OVER (PARTITION BY C.CalendarId
ORDER BY StartDate,EndDate
ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING),
ResourceKey
FROM dbo.Calendar AS C
LEFT JOIN dbo.CalendarEntry
ON CalendarEntry.CalendarId = C.CalendarId;
If you want to use FIRST_VALUE function, then you should rewrite the order by as below:
ORDER BY StartDate DESC,EndDate DESC
And you also you will not need to specify the ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING section
SELECT DISTINCT
Name,
FIRST_VALUE(StartDate) OVER (PARTITION BY C.CalendarId
ORDER BY StartDate DESC,EndDate DESC),
FIRST_VALUE(EndDate) OVER (PARTITION BY C.CalendarId
ORDER BY StartDate DESC,EndDate DESC),
ResourceKey
FROM dbo.Calendar AS C
LEFT JOIN dbo.CalendarEntry
ON CalendarEntry.CalendarId = C.CalendarId;
I have a table including more than 5 million rows of sales transactions. I would like to find sum of date intervals between each customer three recent purchases.
Suppose my table looks like this :
CustomerID ProductID ServiceStartDate ServiceExpiryDate
A X1 2010-01-01 2010-06-01
A X2 2010-08-12 2010-12-30
B X4 2011-10-01 2012-01-15
B X3 2012-04-01 2012-06-01
B X7 2012-08-01 2013-10-01
A X5 2013-01-01 2015-06-01
The Result that I'm looking for may looks like this :
CustomerID IntervalDays
A 802
B 135
I know the query need to first retrieve 3 resent transactions of each customer (based on ServiceStartDate) and then calculate the interval between startDate and ExpiryDate of his/her transactions.
You want to calculate the difference between the previous row's ServiceExpiryDate and the current row's ServiceStartDate based on descending dates and then sum up the last two differences:
with cte as
(
select tab.*,
row_number()
over (partition by customerId
order by ServiceStartDate desc
, ServiceExpiryDate desc -- don't know if this 2nd column is necessary
) as rn
from tab
)
select t2.customerId,
sum(datediff(day, prevEnd, ServiceStartDate)) as Intervaldays
,count(*) as purchases
from cte as t2 left join cte as t1
on t1.customerId = t2.customerId
and t1.rn = t2.rn+1 -- previous and current row
where t2.rn <= 3 -- last three rows
group by t2.customerId;
Same result using LEAD:
with cte as
(
select tab.*,
row_number()
over (partition by customerId
order by ServiceStartDate desc) as rn
,lead(ServiceExpiryDate)
over (partition by customerId
order by ServiceStartDate desc
) as prevEnd
from tab
)
select customerId,
sum(datediff(day, prevEnd, ServiceStartDate)) as Intervaldays
,count(*) as purchases
from cte
where rn <= 3
group by customerId;
Both will not return the expected result unless you subtract purchases (or max(rn)) from Intervaldays. But as you only sum two differences this seems to be not correct for me either...
Additional logic must be applied based on your rules regarding:
customer has less than 3 purchases
overlapping intervals
Assuming there are no overlaps, I think you want this:
select customerId,
sum(datediff(day, ServiceStartDate, ServieEndDate) as Intervaldays
from (select t.*, row_number() over (partition by customerId
order by ServiceStartDate desc) as seqnum
from table t
) t
where seqnum <= 3
group by customerId;
Try this:
SELECT dt.CustomerID,
SUM(DATEDIFF(DAY, dt.PrevExpiry, dt.ServiceStartDate)) As IntervalDays
FROM (
SELECT *
, ROW_NUMBER() OVER (PARTITION BY CustomerID ORDER BY ServiceStartDate DESC) AS rn
, (SELECT Max(ti.ServiceExpiryDate)
FROM yourTable ti
WHERE t.CustomerID = ti.CustomerID
AND ti.ServiceStartDate < t.ServiceStartDate) As PrevExpiry
FROM yourTable t )dt
GROUP BY dt.CustomerID
Result will be:
CustomerId | IntervalDays
-----------+--------------
A | 805
B | 138